Fusion of Intelligent Control and Acoustic Sensing for an Autonomous Helicopter

An autonomous helicopter presents a highly complex system that is difficult to control. This thesis introduces two novel approaches to enhance the control of the autonomous helicopter Flyper and to stabilise its flight: Evolutionary algorithm based control parameter identification and tuning as well as acoustic sensing methodologies. Evolutionary control parameter identification and tuning on a real-world system can provide controllers without the need to identify a formal model. This research provides evidence that natural noise and uncertainties force a genetic algorithm to carry on evolving towards more robust controllers even when a fitness plateau has apparently been reached. It is shown that the introduced methodology can produce robust controllers. The intrinsic sound signature of the helicopter is recorded and analysed using a microphone array connected to a base station. Novel techniques are introduced that extract a multitude of information from the sound signature of the helicopter in flight, without the need for on-board sensors that would add to the payload. A hybrid evolutionary and artificial neural network methodology is shown to be capable of training complex estimators. This technique has been used to measure the individual rotor speeds of the coaxial rotor helicopter from its intrinsic sound signature only. Besides the individual rotor speeds, the helicopter's state and its distance and direction from the microphone array are extracted in real-time. This information is fed back to the helicopter and fused with its controllers to further stabilise its flight. Test flights confirmed that acoustic sensing can significantly enhance the helicopter and its stability in-flight. This work demonstrates that acoustic sensing is a valuable asset that is currently little explored and underused in machine sensing and control.